I recently had a conversation with someone who is very experienced in AI and also hardware and he said this when I said I am interested in solving mechanistic interpretability and I sent him my Optimization Daemons article.
Just a quick response here - please do not present determinism as a fact. The fact is, determinism is one of many different philosophical theories and definitely not an uncontested one.
Furthermore, you argue that you can predict human decisions with 100% accuracy - that is not correct even if determinism was actually correct. This is because quantum uncertainty plays a role in human neural processing and so there will always be, at the very least, a partial randomness element that prevents any 100%-certain predictions. And this is supported by studies, see quantum cognition.
If there is any randomness that doesn’t play on free will argument because we have no control over it
Any other decision that’s being made will be predictable from your brain state right?
Take the simple example of an LLM, we do random sampling from its output distribution, if we remove random and just pick top 1 all the time we can predict what it’s going to say by basically running the model.
Of course there are some GPU issues where there’s still some randomness, why do you think it’s different for our brains?
I’m not saying our brains are static, they change all the time, but if I have a perfect snapshot then I can predict by basically running the same algorithm the brain runs, that’s my argument.
Yes, you are right, quantum (true) randomness on its own does not logically equate to free will - but it does negate the ability to predict decisions with 100% accuracy.
The difference between human brains and current computers is that brains have so much quantum randomness happening constantly that you just cannot make any accurate predictions - not practically neither theoretically. Even if you have a perfect static snapshot, when you make a computation the particles need to move and interact and there is quantum randomness affecting the precise outcomes of these interactions.
Whereas for computers you can make predictions extremely easily even in practice.
That would no longer be the case if you introduced something like a quantum random number generator into LLM token sampling, or deeper within their architecture. Which is something I want to explore with Tetherware.
I need to read more about quantum but, I think we humans are fairly consistent in a small time period, so maybe randomness exists but our brain has some correction capabilities?
Just a quick response here - please do not present determinism as a fact. The fact is, determinism is one of many different philosophical theories and definitely not an uncontested one.
Furthermore, you argue that you can predict human decisions with 100% accuracy - that is not correct even if determinism was actually correct. This is because quantum uncertainty plays a role in human neural processing and so there will always be, at the very least, a partial randomness element that prevents any 100%-certain predictions. And this is supported by studies, see quantum cognition.
My personal favorite in terms of philosophical interpretations of consciousness is currently the quantum information panpsychism. I wholeheartedly recommend checking it out: https://www.essentiafoundation.org/quantum-fields-are-conscious-says-the-inventor-of-the-microprocessor/seeing/
If there is any randomness that doesn’t play on free will argument because we have no control over it
Any other decision that’s being made will be predictable from your brain state right?
Take the simple example of an LLM, we do random sampling from its output distribution, if we remove random and just pick top 1 all the time we can predict what it’s going to say by basically running the model.
Of course there are some GPU issues where there’s still some randomness, why do you think it’s different for our brains?
I’m not saying our brains are static, they change all the time, but if I have a perfect snapshot then I can predict by basically running the same algorithm the brain runs, that’s my argument.
What’s your take on this?
Yes, you are right, quantum (true) randomness on its own does not logically equate to free will - but it does negate the ability to predict decisions with 100% accuracy.
The difference between human brains and current computers is that brains have so much quantum randomness happening constantly that you just cannot make any accurate predictions - not practically neither theoretically. Even if you have a perfect static snapshot, when you make a computation the particles need to move and interact and there is quantum randomness affecting the precise outcomes of these interactions.
Whereas for computers you can make predictions extremely easily even in practice.
That would no longer be the case if you introduced something like a quantum random number generator into LLM token sampling, or deeper within their architecture. Which is something I want to explore with Tetherware.
I need to read more about quantum but, I think we humans are fairly consistent in a small time period, so maybe randomness exists but our brain has some correction capabilities?